FastRNA: An efficient solution for PCA of single-cell RNA-sequencing data based on a batch-accounting count model

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis commonly uses principal component analysis (PCA) on log-transformed data, which can introduce bias.
• Existing count-based PCA methods for scRNA-seq data are computationally intensive and slow for large datasets.
• Standard log normalization also becomes inefficient as scRNA-seq data size increases.

Purpose of the Study:

• To develop a highly efficient PCA method for scRNA-seq data that directly models count data.
• To address the computational limitations of existing PCA methods for large-scale scRNA-seq datasets.
• To account for batch effects and cell size factors within the PCA framework.

Main Methods:

• Developed FastRNA, an efficient PCA solution for scRNA-seq data based on a count model.
• Employs unique algebraic optimization to avoid forming large, dense residual matrices.
• Integrates batch effect correction prior to PCA computation.

Main Results:

• FastRNA achieves two orders of magnitude less time and memory usage compared to other count-based PCA methods.
• It is an order of magnitude more efficient than standard log normalization in terms of time and memory.
• FastRNA can generate batch-corrected principal components for datasets with 2 million cells in under a minute using 1 GB of memory.

Conclusions:

• FastRNA provides a computationally efficient and scalable solution for PCA on scRNA-seq data.
• The method accurately models count data while effectively mitigating batch effects.
• Enables rapid analysis of large-scale scRNA-seq datasets, facilitating deeper biological insights.